Skip to content

Contributing

The contributor guide is CONTRIBUTING.md in the repository. It covers:

  • Getting started: clone, npm install, npm run build, npm test, and npm run hooks:install for the commit hooks.
  • What must never be published: session URLs and a maintainer’s own wallet addresses or SuiNS names, and the hooks that block them.
  • Adding a new tool: argument helpers, profiles, the generated tool reference, store overrides, tests and a live check.
  • Live checks, including oracle-pass, which compares sampled answers (swap labels, object changes, mint authority, flows, cap owners, funding, holders, staking, bridge events, past balances and checkpoint lookups) with the same facts read from the chain by another path, and consistency-pass comparisons of transaction, funding, balance, flow and identity answers on the same subjects. An affected-tier run selects these passes when a tool they check changes; bounded reads are reported separately.
  • Running a blind investigation and writing a case file.
  • Changing an anomaly detector.
  • Keeping the protocol registry and curated verified coins current; uncurated mainnet symbol discovery searches live DEX pools and a bounded metadata scan.
  • Citing the Sui framework source for claims about it.
  • Writing mocks that match what a service can return.
  • Adding a bridge.
  • Weekly data-refresh PRs: pending Unreleased changes receive data under Unreleased for the next normal release, while a data-only PR bumps the patch version and publishes on merge. A maintainer approves the bot PR’s CI run.
  • Releasing, and what needs a new release.

To build and run the server from a clone, see Running from source.

For context-size comparisons, build the server and save a case-pass summary:

Terminal window
npm run build
npm run verify:live -- --keep-summary /tmp/case-pass.json
node scripts/probe/token-baseline.mjs --summary /tmp/case-pass.json --out /tmp/token-baseline

The report measures tool definitions by profile selection and answer sizes from the saved run without repeating its tool calls. It reports characters and estimated tokens. The script is a manual comparison tool, not a verify:live check.